In the SSH protocol, clients and servers execute the key exchange to
generate one-time session keys used for encryption and authentication.
The key exchange is performed initially after the connection is
established and then periodically after a configurable amount of data.
While a key exchange is in progress, we add the received packets to an
internal queue until we receive SSH_MSG_KEXINIT from the other side.
This can result in high memory usage if the other party is slow to
respond to the SSH_MSG_KEXINIT packet, or memory exhaustion if a
malicious client never responds to an SSH_MSG_KEXINIT packet during a
large file transfer.
We now limit the internal queue to 64 packets: this means 2MB with the
typical 32KB packet size.
When the internal queue is full we block further writes until the
pending key exchange is completed or there is a read or write error.
Thanks to Yuichi Watanabe for reporting this issue.
Change-Id: I1ce2214cc16e08b838d4bc346c74c72addafaeec
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/652135
Reviewed-by: Neal Patel <nealpatel@google.com>
Auto-Submit: Gopher Robot <gobot@golang.org>
Reviewed-by: Roland Shoemaker <roland@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Implement the "strict KEX" protocol changes, as described in section
1.9 of the OpenSSH PROTOCOL file (as of OpenSSH version 9.6/9.6p1).
Namely this makes the following changes:
* Both the server and the client add an additional algorithm to the
initial KEXINIT message, indicating support for the strict KEX mode.
* When one side of the connection sees the strict KEX extension
algorithm, the strict KEX mode is enabled for messages originating
from the other side of the connection. If the sequence number for
the side which requested the extension is not 1 (indicating that it
has already received non-KEXINIT packets), the connection is
terminated.
* When strict kex mode is enabled, unexpected messages during the
handshake are considered fatal. Additionally when a key change
occurs (on the receipt of the NEWKEYS message) the message sequence
numbers are reset.
Thanks to Fabian Bäumer, Marcus Brinkmann, and Jörg Schwenk from Ruhr
University Bochum for reporting this issue.
Fixes CVE-2023-48795
Fixesgolang/go#64784
Change-Id: I96b53afd2bd2fb94d2b6f2a46a5dacf325357604
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/550715
Reviewed-by: Nicola Murino <nicola.murino@gmail.com>
Reviewed-by: Tatiana Bradley <tatianabradley@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Roland Shoemaker <roland@golang.org>
Reviewed-by: Damien Neil <dneil@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
This fixes a data race in the tests for x/crypto/ssh, which expects to
be able to examine a transport's read and write counters without
locking after closing it.
(Given the number of goroutines, channels, and mutexes used in this
package, I wouldn't be surprised if other concurrency bugs remain.
I would suggest simplifying the concurrency in this package, but I
don't intend to follow up on that myself at the moment.)
Fixesgolang/go#56957.
Change-Id: Ib1f1390b66707c66a3608e48f3f52483cff3c1f5
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/456758
Reviewed-by: Roland Shoemaker <roland@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Bryan Mills <bcmills@google.com>
Run-TryBot: Bryan Mills <bcmills@google.com>
In accordance to RFC8308, send ext-info-c only during the first key
exchange. Some server implementations such as OpenSSH 7 will send an
extInfoMsg message each time when ext-info-c is received. This results
in a closed connection, as our client does not expect this message while
handling the mux.
See https://bugzilla.mindrot.org/show_bug.cgi?id=2929 regarding the
behaviour of OpenSSH if it sees ext-info-c in later key exchanges.
Fixesgolang/go#51808
Change-Id: Id94f1ef73cec6147136246b0b6048b57db92660d
GitHub-Last-Rev: fcfe5ed37306136219854031abc809e0dc9b3124
GitHub-Pull-Request: golang/crypto#208
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/394134
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Run-TryBot: Filippo Valsorda <filippo@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Trust: Roland Shoemaker <roland@golang.org>
CL 220037 had implemented support for host authentication using
rsa-sha2-256/512, but not client public key authentication. OpenSSH
disabled the SHA-1 based ssh-rsa by default in version 8.8 (after
pre-announcing it in versions 8.2, 8.3, 8.4, 8.5, 8.6, and 8.7) although
some distributions re-enable it. GitHub will start rejecting ssh-rsa for
keys uploaded before November 2, 2021 on March 15, 2022.
https://github.blog/2021-09-01-improving-git-protocol-security-github/
The server side already worked, as long as the client selected one of
the SHA-2 algorithms, because the signature flowed freely to Verify.
There was however nothing verifying that the signature algorithm matched
the advertised one. The comment suggested the check was being performed,
but it got lost back in CL 86190043. Not a security issue because the
signature had to pass the callback's Verify method regardless, and both
values were checked to be acceptable.
Tested with OpenSSH 8.8 configured with "PubkeyAcceptedKeyTypes -ssh-rsa"
and no application-side changes.
The Signers returned by ssh/agent (when backed by an agent client)
didn't actually implement AlgorithmSigner but ParameterizedSigner, an
interface defined in an earlier version of CL 123955.
Updates golang/go#49269Fixesgolang/go#39885
For golang/go#49952
Change-Id: I13b41db8041f1112a70f106c55f077b904b12cb8
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/392394
Trust: Filippo Valsorda <filippo@golang.org>
Run-TryBot: Filippo Valsorda <filippo@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Roland Shoemaker <roland@golang.org>
The server implementation looks at the HostKeys to advertise and
negotiate host key signature algorithms. A fundamental issue of the
Signer and AlgorithmSigner interfaces is that they don't expose the
supported signature algorithms, so really the server has to guess.
Currently, it would guess exclusively based on the PublicKey.Type,
regardless of whether the host key implemented AlgorithmSigner. This
means that a legacy Signer that only supports ssh-rsa still led the
server to negotiate rsa-sha2 algorithms. The server would then fail to
find a suitable host key to make the signature and crash.
This won't happen if only Signers from this package are used, but if a
custom Signer that doesn't support SignWithAlgorithm() but returns
"ssh-rsa" from PublicKey().Type() is used as a HostKey, the server is
vulnerable to DoS.
The only workable rules to determine what to advertise seems to be:
1. a pure Signer will always Sign with the PublicKey.Type
2. an AlgorithmSigner supports all algorithms associated with the
PublicKey.Type
Rule number two means that we can't add new supported algorithms in the
future, which is not great, but it's too late to fix that.
rsaSigner was breaking rule number one, and although it would have been
fine where it's used, I didn't want to break our own interface contract.
It's unclear why we had separate test key entries for rsa-sha2
algorithms, since we can use the ssh-rsa key for those. The only test
that used them, TestCertTypes, seemed broken: the init was actually
failing at making the corresponding signers rsaSigners, and indeed the
test for the SHA-256 signer expected and checked a SHA-512 signature.
Pending CVE
For golang/go#49952
Change-Id: Ie658eefcadd87906e63fc7faae8249376aa96c79
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/392355
Trust: Filippo Valsorda <filippo@golang.org>
Run-TryBot: Filippo Valsorda <filippo@golang.org>
Reviewed-by: Roland Shoemaker <roland@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
RFC 8332, Section 2 sets up two overlapping namespaces: public key
formats and public key algorithms.
* The formats are what we currently have KeyAlgo constants for, and they
appear in PublicKey.Type.
* The algorithms are the set of both KeyAlgo and SigAlgo constants, and
they appear in Signature.Format (amongst other places).
This is incoherent, because that means Signature.Format can be both a
KeyAlgo (like KeyAlgoECDSA256) or a SigAlgo (like SigAlgoRSASHA2256).
One solution would be to duplicate all the KeyAlgo constants into the
SigAlgo namespace, but that would be confusing because applications are
currently using KeyAlgos where they'd be supposed to use the new
SigAlgos (while we can't deprecate the KeyAlgos because they are still
necessary for the PublicKey.Type namespace).
Instead, drop the separate namespaces, and use KeyAlgos throughout.
There are simply some KeyAlgos that can't be a PublicKey.Type.
Take the opportunity to fix the stuttering SHA22565/SHA2512 names. It's
totally ok to call those hashes SHA-256 and SHA-512 without the family
infix.
For golang/go#49952
Change-Id: Ia1fce3912a7e60aa70a88f75ed311be331fd19d5
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/392354
Trust: Filippo Valsorda <filippo@golang.org>
Run-TryBot: Filippo Valsorda <filippo@golang.org>
Reviewed-by: Roland Shoemaker <roland@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
This change adds support for RSA SHA-2 based signatures for host keys and certificates. It also switches the default certificate signature algorithm for RSA to use SHA-512. This is implemented by treating ssh.Signer specially when the key type is `ssh-rsa` by also allowing SHA-256 and SHA-512 signatures.
Fixesgolang/go#37278
Change-Id: I2ee1ac4ae4c9c1de441a2d6cf1e806357ef18910
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/220037
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Jason A. Donenfeld <Jason@zx2c4.com>
Reviewed-by: Roland Shoemaker <roland@golang.org>
At the protocol level, SSH lets client and server specify different
algorithms for the read and write half of the connection. This has
never worked correctly, as Client-to-Server was always interpreted as
the "write" side, even if we were the server.
This has never been a problem because, apparently, there are no
clients that insist on different algorithm choices running against Go
SSH servers.
Since the SSH package does not expose a mechanism to specify
algorithms for read/write separately, there is end-to-end for this
change, so add a unittest instead.
Change-Id: Ie3aa781630a3bb7a3b0e3754cb67b3ce12581544
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/172538
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Run-TryBot: Filippo Valsorda <filippo@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
According to RFC 4252 section 5.4, the banner is sent between the
ssh-connection request and responding to user authentication.
Original support for server sending banner by joshua stein <jcs@jcs.org>
Fixesgolang/go#19567
Change-Id: I729b3c8e5fd2c0068609d1590b61e92f40d87ea4
Reviewed-on: https://go-review.googlesource.com/71790
Run-TryBot: Han-Wen Nienhuys <hanwen@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Han-Wen Nienhuys <hanwen@google.com>
According to RFC 4252 section 5.4, the banner is sent between the
ssh-connection request and responding to user authentication.
Original support for server sending banner by joshua stein <jcs@jcs.org>
Fixesgolang/go#19567
Change-Id: I68944a7f4711c0623759f6a59023e8e45a8781aa
Reviewed-on: https://go-review.googlesource.com/65271
Reviewed-by: Han-Wen Nienhuys <hanwen@google.com>
Run-TryBot: Han-Wen Nienhuys <hanwen@google.com>
Fixes a nil pointer dereference that slipped through buildbots because
it was introduced by the last two commits.
Change-Id: Ib269e910956cd8b3b46e217b03fde1b61572260a
Reviewed-on: https://go-review.googlesource.com/40530
Reviewed-by: Han-Wen Nienhuys <hanwen@google.com>
Run-TryBot: Han-Wen Nienhuys <hanwen@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
The normal handshake kicks off with a waitSession(), which guarantees
that we never attempt to send data before the first kex is completed,
but ensuring readPacketsLeft > 0 and writePacketsLeft > 0 helps
understand that thresholds can never cause spurious rekeying at the
start of a connection.
Change-Id: If5bcafcda0c7d16fd21f22c664101ac5f5b487d7
Reviewed-on: https://go-review.googlesource.com/38696
Reviewed-by: Adam Langley <agl@golang.org>
Run-TryBot: Adam Langley <agl@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This change breaks existing behavior.
Before, a missing ClientConfig.HostKeyCallback would cause host key
checking to be disabled. In this configuration, establishing a
connection to any host just works, so today, most SSH client code in
the wild does not perform any host key checks.
This makes it easy to perform a MITM attack:
* SSH installations that use keyboard-interactive or password
authentication can be attacked with MITM, thereby stealing
passwords.
* Clients that use public-key authentication with agent forwarding are
also vulnerable: the MITM server could allow the login to succeed, and
then immediately ask the agent to authenticate the login to the real
server.
* Clients that use public-key authentication without agent forwarding
are harder to attack unnoticedly: an attacker cannot authenticate the
login to the real server, so it cannot in general present a convincing
server to the victim.
Now, a missing HostKeyCallback will cause the handshake to fail. This
change also provides InsecureIgnoreHostKey() and FixedHostKey(key) as
ready made host checkers.
A simplistic parser for OpenSSH's known_hosts file is given as an
example. This change does not provide a full-fledged parser, as it
has complexity (wildcards, revocation, hashed addresses) that will
need further consideration.
When introduced, the host checking feature maintained backward
compatibility at the expense of security. We have decided this is not
the right tradeoff for the SSH library.
Fixesgolang/go#19767
Change-Id: I45fc7ba9bd1ea29c31ec23f115cdbab99913e814
Reviewed-on: https://go-review.googlesource.com/38701
Run-TryBot: Han-Wen Nienhuys <hanwen@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
The previous attempt would fail in the following scenario:
* select picks "first" kex from requestKex
* read loop receives a remote kex, posts on requestKex (which is now
empty) [*] for sending out a response, and sends pendingKex on startKex.
* select picks pendingKex from startKex, and proceeds to run the key
exchange.
* the posting on requestKex in [*] now triggers a second key exchange.
Fixes#18861.
Change-Id: I443e82f1d04c7f17d1485fdb87072b9feec26aa8
Reviewed-on: https://go-review.googlesource.com/36055
Run-TryBot: Han-Wen Nienhuys <hanwen@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Han-Wen Nienhuys <hanwen@google.com>
Since encryption messes up the packets, the wrongly retained packets
look like noise and cause application protocol errors or panics in the
SSH library.
This normally triggers very rarely: the mandatory key exchange doesn't
have parallel writes, so this failure condition would be setup on the
first key exchange, take effect only after the second key exchange.
Fortunately, the tests against openssh exercise this. This change adds
also adds a unittest.
Fixes#18850.
Change-Id: I656c8b94bfb265831daa118f4d614a2f0c65d2af
Reviewed-on: https://go-review.googlesource.com/36056
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Han-Wen Nienhuys <hanwen@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
1) Always force a key exchange if we exchange 2^31 packets. In the past
this might not happen if RekeyThreshold was set to a very large
interval.
2) Follow recommendations from RFC 4344 for block ciphers. For AES, we
can encrypt 2^(blocksize/4) blocks under the same keys.
On modern hardware, the previous default of 1Gb could force a key
exchange within ~10 seconds. Since the key exchange takes 3 roundtrips
(send kex init, send DH init, send NEW_KEYS), this is relatively
expensive on high-latency links.
Change-Id: I1297124a307c541b7bf22d814d136ec0c6d8ed97
Reviewed-on: https://go-review.googlesource.com/35410
Run-TryBot: Han-Wen Nienhuys <hanwen@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Adam Langley <agl@golang.org>
The initial kex is started from both sides simultaneously, and before,
we could consume the the incoming kex request before we consumed from
our internal channel. This would result in initiating a key exchange
just after completing the initial one, which is not only an extra
delay, but also an error when using OpenSSH (OpenSSH does not support
key exchanges during user authentication).
Change-Id: Ia7e0748ea2bca80ae97d187bcf2931ab6422276b
Reviewed-on: https://go-review.googlesource.com/35851
Run-TryBot: Han-Wen Nienhuys <hanwen@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Adam Langley <agl@golang.org>
Use channels and a dedicated write loop for managing the rekeying
process. This lets us collect packets to be written while a key
exchange is in progress.
Previously, the read loop ran the key exchange, and writers would
block if a key exchange was going on. If a reader wrote back a packet
while processing a read packet, it could block, stopping the read
loop, thus causing a deadlock. Such coupled read/writes are inherent
with handling requests that want a response (eg. keepalive,
opening/closing channels etc.). The buffered channels (most channels
have capacity 16) papered over these problems, but under load SSH
connections would occasionally deadlock.
Fixes#18439.
Change-Id: I7c14ff4991fa3100a5d36025125d0cf1119c471d
Reviewed-on: https://go-review.googlesource.com/35012
Run-TryBot: Han-Wen Nienhuys <hanwen@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Han-Wen Nienhuys <hanwen@google.com>
In the initial key exchange, a client has the option of sending a
guessed key exchange packet. This guess is considered wrong (RFC 4253
Section 7) if "the kex algorithm and/or the host key algorithm is
guessed wrong (server and client have different preferred algorithm),
or if any of the other algorithms cannot be agreed upon...".
The library should be checking the first algorithm in the supported
algorithms, not the agreed algorithm. It also needs to check both the
kex algorithm and the host key algorithm.
Fixesgolang/go#16962
Change-Id: I6b62b1f5b39731326280571d373635085135a2a1
Reviewed-on: https://go-review.googlesource.com/28750
Reviewed-by: Han-Wen Nienhuys <hanwen@google.com>
Run-TryBot: Han-Wen Nienhuys <hanwen@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
If one of both sides is slow, the first kex completes implicitly. The
first kex also produces msgNewKeys, and this must be read to ensure
that the authentication code is not confused by it.
Before this fix, the problem could be reproduced by inserting a sleep
just before the requestInitialKeyChange call.
Fixes#15198
Change-Id: I602db5dd37b2d8556c88ab4cdb693ccf90147a3d
Reviewed-on: https://go-review.googlesource.com/23137
Run-TryBot: Han-Wen Nienhuys <hanwen@google.com>
Reviewed-by: Adam Langley <agl@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This is done by running the key exchange and setting the session ID
under mutex. If the first exchange encounters an already set session
ID, then do nothing.
This fixes a race condition:
On setting up the connection, both sides sent a kexInit to initiate
the first (mandatory) key exchange. If one side was faster, the
faster side might have completed the key exchange, before the slow
side had a chance to send a kexInit. The slow side would send a
kexInit which would trigger a second key exchange. The resulting
confirmation message (msgNewKeys) would confuse the authentication
loop.
This fix removes sessionID from the transport struct.
This fix also deletes the unused interface rekeyingTransport.
Fixes#15066
Change-Id: I7f303bce5d3214c9bdd58f52d21178a185871d90
Reviewed-on: https://go-review.googlesource.com/21606
Reviewed-by: Adam Langley <agl@golang.org>
Reviewed-by: Han-Wen Nienhuys <hanwen@google.com>
The error message reported by the ssh client when it can't find a
"cipher" in common between the client and server was overly vague. This
adds more detailed error messages to findAgreedAlgorithms so that the
user can more easily identify which of the components can't reach
agreement.
Change-Id: I4d985e92fea964793213e5600b52b3141e712000
Reviewed-on: https://go-review.googlesource.com/13817
Reviewed-by: Adam Langley <agl@golang.org>
Unblock writers if a read error occurs while writers are blocked on a
pending key change.
Add test to check for deadlocks in error paths in handshake.go
Fixesgolang/go#11992.
Change-Id: Id098bd9fec3d4fe83daeb2b7f935e5647c19afd3
Reviewed-on: https://go-review.googlesource.com/13594
Reviewed-by: Adam Langley <agl@golang.org>
Fixesgolang/go#11882
If an error occurs during handshakeTransport.writePacket the lock may not be
released. Fix this by using defer rather than manually unlocking in all paths.
Change-Id: I0010284b4f7d99907c86b4c0e140ab6cf37b0441
Reviewed-on: https://go-review.googlesource.com/12888
Reviewed-by: Adam Langley <agl@golang.org>
Nicola Murino
Brad Fitzpatrick
Roland Shoemaker
Bryan C. Mills
Peter Verraedt
Hans Nielsen
Han-Wen Nienhuys
Tugdual Saunier
Kaleb Elwert
Thomas Desrosiers
Dave Cheney
Adam Langley